HEAT CONVECTION ENHANCEMENT OF UNILATERAL-HEATED SQUARE CHANNELS BY INCLINED RIBS OPTIMIZED BY MACHINE LEARNING

Optimizing structure parameters is pivotal in enhancing convective heat. This study leverages machine learning methods to establish a relationship between input parameters and targets, providing a novel approach to structure parameter optimization in convective heat transfer of a unilateral-heated square channel with inclined ribs. A substantial dataset is procured through batch modeling and computational fluid dynamics (CFD) simulations by changing the rib structure (length, diameter, and spacing). The Gaussian process regression is applied to train the surrogate model due to its continuity and smoothness. Two scenarios are considered: the datum of the best structure from CFD simulation is included or excluded in the training dataset. When the datum from the best CFD structure is included in the training dataset, the optimized Nusselt number and the comprehensive heat transfer characteristic performance evaluation criterion (PEC) from the surrogate model are almost the same with the best CFD result. When it is excluded, the surrogate model optimization can yield a 7% and 3% increase in Nu and PEC. The heat transfer mechanism and heat transfer effects of the unilateral-heated square channels with inclined ribs are analyzed. The optimal Nu and PEC under different Reynolds numbers are obtained, and the variation of geometric parameters with the optimal solution is analyzed. This study underscores the potential of machine learning in optimizing convective heat transfer channels, benefiting future research and applications in this field.